Activized Learning: Transforming Passive to Active with Improved Label Complexity

TL;DR

This paper demonstrates that active learning can significantly reduce label complexity compared to passive learning for VC class classifiers, even with label noise, through theoretical analysis and novel metrics.

Contribution

It provides a theoretical framework showing how active learning outperforms passive learning in label efficiency, introducing a generalized disagreement coefficient.

Findings

Active learning achieves asymptotically better label complexity in noise-free settings.

The generalized disagreement coefficient characterizes the magnitude of improvements.

Strict improvements over passive learning are possible even with label noise.

Abstract

We study the theoretical advantages of active learning over passive learning. Specifically, we prove that, in noise-free classifier learning for VC classes, any passive learning algorithm can be transformed into an active learning algorithm with asymptotically strictly superior label complexity for all nontrivial target functions and distributions. We further provide a general characterization of the magnitudes of these improvements in terms of a novel generalization of the disagreement coefficient. We also extend these results to active learning in the presence of label noise, and find that even under broad classes of noise distributions, we can typically guarantee strict improvements over the known results for passive learning.